Dawkins’ WEASEL: Proximity Search With or Without Locking?

_{William Dembski
March 16, 2009

Darwinism, Evolution, Informatics

3}_{Categories
Darwinism
Evolution
Informatics}

Share: Facebook; Twitter/X; LinkedIn; Flipboard; Print; Email

On pp. 47-48 of THE BLIND WATCHMAKER, Richard Dawkins gives two runs of his WEASEL program (note that there were typos in both initial seeds — one had 27 characters, the other 29 whereas they should have 28; I’ve corrected that). Here are the two runs using the Courier typeface, which assigns equal width to each character (spaces are represented by asterisks):

WDL*MNLT*DTJBKWIRZREZLMQCO*P
WDLTMNLT*DTJBSWIRZREZLMQCO*P
MDLDMNLS*ITJISWHRZREZ*MECS*P
MELDINLS*IT*ISWPRKE*Z*WECSEL
METHINGS*IT*ISWLIKE*B*WECSEL
METHINKS*IT*IS*LIKE*I*WEASEL
METHINKS*IT*IS*LIKE*A*WEASEL

Y*YVMQKZPFJXWVHGLAWFVCHQXYPY
Y*YVMQKSPFTXWSHLIKEFV*HQYSPY
YETHINKSPITXISHLIKEFA*WQYSEY
METHINKS*IT*ISSLIKE*A*WEFSEY
METHINKS*IT*ISBLIKE*A*WEASES
METHINKS*IT*ISJLIKE*A*WEASEO
METHINKS*IT*IS*LIKE*A*WEASEP
METHINKS*IT*IS*LIKE*A*WEASEL

These runs are incomplete. The first, according to Dawkins, required 43 iterations to converge, the second 64 (Dawkins omitted the other iterates to save space).

As you can see, by using the Courier font, one can read up from the target sequence METHINKS*IT*IS*LIKE*A*WEASEL, as it were column by column, over each letter of the target sequence. From this it’s clear that once the right letter in the target sequence is latched on to, it locks on and never changes. In other words, in these examples of Dawkins’ WEASEL program as given in his book THE BLIND WATCHMAKER, it never happens (as far as we can tell) that some intermediate sequences achieves the corresponding letter in the target sequence, then loses it, and in the end regains it.

Thus, since Dawkins does not make explicit in THE BLIND WATCHMAKER just how his algorithm works, it is natural to conclude that it is a proximity search with locking (i.e., it locks on characters in the target sequence and never lets go).

Interestingly, when Dawkins did his 1987 BBC Horizons takeoff on his book, he ran the program in front of the film camera:

www.youtube.com/watch?v=5sUQIpFajsg (go to 6:15)

There you see that his WEASEL program does a proximity search without locking (letters in the target sequence appear, disappear, and then reappear).

That leads one to wonder whether the WEASEL program, as Dawkins had programmed and described it in his book, is the same as in the BBC Horizons documentary.

In any case, our chief programmer at the Evolutionary Informatics Lab (www.evoinfo.org) is expanding our WEASEL WARE software to model both these possibilities. Stay tuned.

Comments

R0b, KF, hazel, gpuccio, joseph, Patrick, et al., I finished coding Weasel Ware 2.0 this weekend, with lots of new features and goodies. I still have some testing to do and such, but one of the features I added was the ability for users to create their own fitness functions, via interpreted javascript. So we can see exactly how much active information a Proximity Reward fitness matrix contributes to the search and users can replace this with a reward matrix generated by a fitness function of their choice. I expect a lot of new discussion to open up once this is released. Oh, and R0b, you now will have both latching and non-latching versions side-by-side, for whichever you want to assume that Dawkins used for The Blind Watchmaker. AtomAtom_{March 23, 2009
March
03
Mar
23
23
2009
11:38 AM
11
11
38
AM
PDT}

Nice post, R0b. It's clear that you understood all this from the beginning, but I learned some things thinking through them myself. I would add that our intuition is also often wrong when it comes to probability. Probability is a complicated subject and leads to many correct but counter-intuitive results.hazel_{March 23, 2009
March
03
Mar
23
23
2009
11:16 AM
11
11
16
AM
PDT}

kairosfocus:
On the evidence of the o/p circa 1986, the simplest explanation is explicit latching of letters once they go correct.
How is the hypothesis that Dawkins added latching simpler than the hypothesis that he didn't add latching?
The 1987 o/p does not latch, and indeed seems to have a puzzling wink and get back pattern.
This is explained by noting that the display cycles through all of the candidate strings in the population, not just the winner.
Bottomline: It is high time Weasel and kin were retired, as serving only rhetorical not illuminative, purposes.
I disagree that it's not illuminative. It shows how the results of the oracle (or environment) simply communicating a fitness level can fool some people into thinking that more information is being communicated. Dembski, kairosfocus, and others thought that the oracle must not just be communicating the number of correct letters, but also which letters are correct. If nothing else, it shows that intuition isn't reliable when it comes to genetic behavior. Note that the artificiality of the fitness measure -- i.e. the fitness measure has nothing to do with any observable functionality -- has never been in dispute. And I agree that Weasel and kin should be retired. Tell that to Marks and Dembski.R0b_{March 23, 2009
March
03
Mar
23
23
2009
10:57 AM
10
10
57
AM
PDT}

Joseph [223], I meant ratcheting in the sense defined by hazel. I'll just refer to hazel [227] for clarification. kairosfocus [229], I neither know nor care what you are talking about in your PS. UB [232], Berlinski has a Ph.D. in philosophy and has done some work in mathematics but hasn't contributed to the mathematical literature as far as I know. That's what I meant. I would bet his mental faculties are as fine, or not, as ever. My point was that his essay does not address the mathematics of Dawkins's program (the immediate point of discussion) or even its claims; it just makes an interpretation as to its value.David Kellogg_{March 23, 2009
March
03
Mar
23
23
2009
10:13 AM
10
10
13
AM
PDT}

Kellog at 202 I'm not sure on what grounds one could insiuate that David's mathmatical acumen is somehow deteriorated. You are correct however that his analysis was not pure mathematics, but, I am quite certain that was his specific intent. He makes the larger point clear, there is a target in Dawkin's exercise. The simulation reaches a predefined goal. Life, or more appropriately evolution, does not work that way. There is no target. Produce "methinks its a weasel" without that phrase existing anywhere in the system, then that will be impressive. Unitl then, Berlinski is 100% correct.Upright BiPed_{March 23, 2009
March
03
Mar
23
23
2009
09:52 AM
9
09
52
AM
PDT}

Patrick, In essence, if ratcheting is implemented, isn't that what is going on? That is with respect to parent and offspring. IOW once a "parent" gets a matching letter, its offspring will also have those matching letters. It is just a product of the process. IOW once the letters are matched the search for them is over- ratchet clicks and fade...Joseph_{March 23, 2009
March
03
Mar
23
23
2009
08:46 AM
8
08
46
AM
PDT}

kf@229, I agree with your first two paragraphs, so I think we can consider the discussion finished. I've never been interested in any of the other issues you mention, though, so I have no comment there.hazel_{March 23, 2009
March
03
Mar
23
23
2009
07:56 AM
7
07
56
AM
PDT}

Hazel: On the evidence of the o/p circa 1986, the simplest explanation is explicit latching of letters once they go correct. Implicit latching and/or quasi-latching is also possible, and that was noted before this thread ever began. In the D-M paper, they looked at the former case, which is a legitimate case of Weasel, given the many versions floating out there. The 1987 o/p does not latch, and indeed seems to have a puzzling wink and get back pattern. On preponderance of evidence, including Mr Dawkins reported claim that he did not explicitly latch, the program, implicit latching as has been outlined many times above, is a reasonable explanation. In either case, latching or quasi latching are secondary to Weasel's real downfall: it is targetted search that rewards proximity without functionality. So, it is not a reasonable part of any case pointing to a BLIND watchmaker; the presumable aim of the book, BW. That point was made ever since December last, but it seems that raising a secondary debate over what is highly evident form the 1986 o/p is highly distractive from that core flaw. Bottomline: It is high time Weasel and kin were retired, as serving only rhetorical not illuminative, purposes. GEM of TKI ________________ PS: Seversky, Kellogg and co associated with Anti Evo, I think you need to address the citation with evident approval here and also at Anti Evo, of someone indulging in, frankly, anti-Christian blood slander and through such ad hominems, dismissal of legitimate concerns over public lewdness. (As I recall, some Dancehall entertainers have actually been censured by the Jamaican courts over their beyond belief on-stage behaviour.)kairosfocus_{March 23, 2009
March
03
Mar
23
23
2009
07:48 AM
7
07
48
AM
PDT}

Thanks for the summary, Patrick.hazel_{March 23, 2009
March
03
Mar
23
23
2009
07:33 AM
7
07
33
AM
PDT}

Joseph, the example about Weasel in the paper you cited shows clearly that Dembski's assumed that there was an explicit rule that locks letters in place. Here's what he wrote:
Partitioned search [12] is a “divide and conquer” procedure best introduced by example. Consider the L = 28 character phrase METHINKS*IT*IS*LIKE*A*WEASEL (19) Suppose the result of our ?rst query of L = 28 characters is SCITAMROFNI*YRANOITULOVE*SAM (20) Two of the letters, {E,S}, are in the correct position. They are shown in a bold font. In partitioned search, our search for these letters is ?nished. For the incorrect letters, we select 26 new letters and obtain OOT*DENGISEDESEHT*ERA*NETSIL (21) Five new letters are found bringing the cumulative tally of discovered characters to {T,S,E,*,E,S,L}. All seven characters are ratcheted into place. Nineteen new letters are chosen and the process is repeated until the entire target phrase is found.
Notice that when you get two correct letters, you only "mutate" the remaining 26, and once you get five more correct, you only mutate the remaining 19. This is an explicit rule. It's is NOT quasi-latching or implicit latching or latching because it is highly improbable that a correct letter mutated to incorrect will get passed on. It is a RULE that says once we have found a letter it is not even considered for further mutation. Period. This is absolutely clear.hazel_{March 23, 2009
March
03
Mar
23
23
2009
07:29 AM
7
07
29
AM
PDT}

Joseph,
I know you can never prove that regression can occur given the parameters I stated.
But it's your claim that regression can never happen. It's up to you to demonstrate this.madsen_{March 23, 2009
March
03
Mar
23
23
2009
07:25 AM
7
07
25
AM
PDT}

I wrote comment #178 because I noticed people seemed to be generally agreeing in principle but yet--not realizing this--were still arguing, apparently due to objecting to certain terminology and descriptors. Never mind personal grievances. To finish this thread, I'll quickly do away with comment #212: Hazel, you know you're taking what Dembski said out of context. Dembski just noted that based upon ONLY the information in the book that it's reasonable to presume a hard-coded latching function. The BBC video makes it obvious that this is not so, so he raises the possibility there are multiple versions of the program and that the issue should be investigated. As in, Dembski is leaving with a question, not an assertion. But given that you and several others have already found an explicit latching function is unnecessary given fine-tuned conditions*, and that Dawkins specifically says that explicit latching was not used, the best explanation at this point is that the program in the book and in the video were likely the same. *Although as kf notes it's possible to get the same results with multiple other approaches. I think that about finishes this thread.Patrick_{March 23, 2009
March
03
Mar
23
23
2009
07:16 AM
7
07
16
AM
PDT}

madsen
While I accept that by adjusting the mutation rate and perhaps some of the other parameters, you can arrange that the probability of regression will be arbitrarily small, I don’t think you will be able to prove it will “NEVER” happen.
I know you can never prove that regression can occur given the parameters I stated. As I said given a small enough mutation rate and a large enough sample size the "worse" that can happen is the output = input. But you could prove me wrong by going to one of the sites tat has the program and running it over and over again, until you find a regression.Joseph_{March 23, 2009
March
03
Mar
23
23
2009
07:01 AM
7
07
01
AM
PDT}

David Kellogg:
the ratcheting process has never been denied by anybody associated with Weasel.
Yes it has:
All seven characters are ratcheted into place.- Dembski/ Marks
IOW "rachet"ing was stated in the paper that you guys are trying to derail. Read it for yourself: Conservation of Information in Search: Measuring the Cost of Success page 5 That said it looks like they were correct as ratcheting does take place as described by Dawkins.Joseph_{March 23, 2009
March
03
Mar
23
23
2009
06:57 AM
6
06
57
AM
PDT}

Joseph,
madsen, If you are REALLY interested just do it- That means if you try you could find the exact parameters in which regression never takes place. Ya see if the mutation rate is small enough it would mean that only one character could change- mutate. It would also mean that there would be exact copies of the parent. So if there are exact copies and mutated copies then AT A MINIMUM, the exact copies would be chosen. Pretty basic actually.
While I accept that by adjusting the mutation rate and perhaps some of the other parameters, you can arrange that the probability of regression will be arbitrarily small, I don't think you will be able to prove it will "NEVER" happen. If you can present a proof, though, I'd be interested in seeing it.madsen_{March 23, 2009
March
03
Mar
23
23
2009
06:56 AM
6
06
56
AM
PDT}

madsen, If you are REALLY interested just do it- That means if you try you could find the exact parameters in which regression never takes place. Ya see if the mutation rate is small enough it would mean that only one character could change- mutate. It would also mean that there would be exact copies of the parent. So if there are exact copies and mutated copies then AT A MINIMUM, the exact copies would be chosen. Pretty basic actually.Joseph_{March 23, 2009
March
03
Mar
23
23
2009
06:47 AM
6
06
47
AM
PDT}

hazel:
I was responding to Dembski’s opening post, in which he referred to a partitioned search, “i.e., it locks on characters in the target sequence and never lets go)."
And that is what happens as described by Dawkins in TBW. That is what happens with ratcheting.Joseph_{March 23, 2009
March
03
Mar
23
23
2009
06:43 AM
6
06
43
AM
PDT}

Now that I've gotten involved, I've gone back and read the start of the thread. I see that ROb made the same point I am making in post 5. All I did was supply a little math that shows why no latching rule is necessary, responding mainly to Dembski's original point that it was "natural" to assume that a latching rule was in place. That's all. Joseph writes,
As I have already stated the program does not need a coded statement for something it does on its own. I take it you cannot understand that. Not my problem…
I was responding to Dembski's opening post, in which he referred to a partitioned search, "i.e., it locks on characters in the target sequence and never lets go)." I was not responding to you. If you accept that there is no latching rule in the algorithm, then we agree.hazel_{March 23, 2009
March
03
Mar
23
23
2009
06:37 AM
6
06
37
AM
PDT}

Joseph,
Given a target, a small enough mutations rate AND a large enough sample size, there will NEVER be any regression. NEVER.
Please prove this mathematically.madsen_{March 23, 2009
March
03
Mar
23
23
2009
06:07 AM
6
06
07
AM
PDT}

There is a big difference, then, between cumulative selection (in which each improvement, however slight, is used as a basis for future building), and single-step selection (in which each new ‘try’ is a fresh one).- Dawkins TBW page 49
That’s ratcheting.Joseph_{March 23, 2009
March
03
Mar
23
23
2009
06:01 AM
6
06
01
AM
PDT}

Sal Gal:
Why Dembski is going on about “locking,” I have no idea.
It's like this- Marks and Dembski wrote a paper. In that paper they referenced "The Blind Watchmaker" pertaing to the "weasel" program Dawkins describes in it. In their paper they used it as an example of a partitioned search- meaning once the letters matched the target they get "locked into place". The anti-IDists have made a big to-do about it. However when looking at ALL the data is is obvious that the program used and described by Dawkins in TBW uses a ratcheting process. And that ratcheting process locks the matched letters in place.Joseph_{March 23, 2009
March
03
Mar
23
23
2009
05:58 AM
5
05
58
AM
PDT}

hazel:
And this should lay to rest, I think, the question of whether the BWM algorithm contained a specific rule about latching, or not.
As I have already stated the program does not need a coded statement for something it does on its own. I take it you cannot understand that. Not my problem...Joseph_{March 23, 2009
March
03
Mar
23
23
2009
05:53 AM
5
05
53
AM
PDT}

madsen:
Looking back through the thread and reading your latest comment, I think we actually agree on what is happening with the program—there isn’t anything in it which prohibits backward movement, but the probability of that happening given the parameters involved is very small.
Given a target, a small enough mutations rate AND a large enough sample size, there will NEVER be any regression. NEVER. IOW ratcheting toward the target, ie locking the correct letters in place, is inevitable.Joseph_{March 23, 2009
March
03
Mar
23
23
2009
05:51 AM
5
05
51
AM
PDT}

hazel: On that the issue is that Weasel circa 1986 and circa 1987 have significantly diverse output characteristics. Shifting the parameters controlling mutation rate and generation size would account for that, and for the purposes of our discussion, would be materially different. GEM of TKIkairosfocus_{March 23, 2009
March
03
Mar
23
23
2009
05:38 AM
5
05
38
AM
PDT}

to David Kellogg: thanks for pointing out that each parent has multiple children. I had never thought about this Weasel program until yesterday when I got interested in the math, so I didn't know that. But I understand now some aspects that weren't clear to me before. Of course this just makes my case stronger. And this should lay to rest, I think, the question of whether the BWM algorithm contained a specific rule about latching, or not. My conclusion is that it did not, and that the BWM algorithm and the video algorithm are the same. This was the issue in Demski's opening post:
Thus, since Dawkins does not make explicit in THE BLIND WATCHMAKER just how his algorithm works, it is natural to conclude that it is a proximity search with locking (i.e., it locks on characters in the target sequence and never lets go).
I disagreed with Dembski, and I think I’ve proved my point.hazel_{March 23, 2009
March
03
Mar
23
23
2009
05:07 AM
5
05
07
AM
PDT}

SG: Weasel's downfall is that it implements acknowledged targetted search that rewards NON-FUNCTIONAL "nonsense" phrases on mere proximity to target. This is NOT a matter of increments in current fitness being rewarded, in any sense of "fitness" worth using. For, Weasel begs the question of achieving significantly complex information based functionality, in its leap to increment proximity of latest generation to target by selection off nearness to target for explicitly non-functional phrases. Thus, we see highlighted the key issue raised by two forms of the design inference: complex funcuoinality based on information is hard to find by random search, and irreducible complexity by direct means or by adaptation of existing parts to form a new function, is also hard to do by the engine of variation, chance. (And, without first having function, on pain of question begging, we cannot see nature culling based on better or worse degrees of function leading to differential success and reproduction.) Further to this, in the recent Marks-Dembski paper, the T2, explicit latch version of the Weasel family of algorithms and implementations is addressed. Searching for target zones and/or target points is . . . search in configuration spaces. (That is, this thread is over an incidental issue, whether in Weasel circa 1986 the correct letters to date are explicitly or implicitly latched, i.e locked or nearly locked in place. A glance at the Dawkins published runs of 1986 shows that once a letter is correct, it seems to be latched in place as Weasel marches on towards its target.) Markov processes and evolutionary fitness strategies may be interesting but they are on a tangent to the key issue here. GEM of TKI PS: Onlookers: the above by SG and my recent response to Dr S in the mod pol thread, underscore how Weasel often succeeds rhetorically by such misdirection while failing to address the issue of increment in complex, functional information to get TO shorelines of islands of function. (The specific issue raised by Hole at eh beginning of the issue, and to which Weasel was claimed to respond.)kairosfocus_{March 23, 2009
March
03
Mar
23
23
2009
04:22 AM
4
04
22
AM
PDT}

By the way, a (1,n)-ES is not a search algorithm in the sense of Dembski and Marks. I see no way to say anything about the active information of a "comma" ES without augmenting the ES with an extrinsic entity that registers the fittest parent of all generations. Of course, then the active information is not measured on the ES, but on the augmented ES. It seems to me that the analytic framework of Dembski and Marks does not apply to the Weasel program -- unless the framework is actually a Procrustean bed.Sal Gal_{March 23, 2009
March
03
Mar
23
23
2009
02:27 AM
2
02
27
AM
PDT}

I cannot believe this thread is still going. Dawkins described what is known in the evolutionary computation community as a (1,n)-ES, where the ES stands for evolution strategy. I informed one of Dembski's friends of this in email dated July 31, 2008. In a (1,n)-ES, 1 parent generates n offspring in each generation, and the parent of the next generation is the fittest of the n offspring in the present generation. In a (1+n)-ES, the parent of the next generation is the fittest of the n+1 individuals in the present generation. That is, in the "comma" strategy, the parent "dies" after generating offspring, and there is a loose analogy to annual plants. In the "plus" strategy, there is a loose analogy to perennials. There is a large body of formal analysis of evolution strategies. I believe that most of it involves Markov chain analysis. Some very bright people have already answered many fundamental questions about the behavior of evolution strategies. There is no need to reinvent the wheel. Why Dembski is going on about "locking," I have no idea. Until recently, the Big Three of evolutionary computation were evolutionary programming, evolution strategies, and genetic algorithms (including genetic programming). I find it difficult to believe that a critic of EC such as Dembski does not recognize an evolution strategy when he sees one. At any rate, Google and ye shall find.Sal Gal_{March 23, 2009
March
03
Mar
23
23
2009
02:10 AM
2
02
10
AM
PDT}

5] I will show why a correct letter mutating to incorrect is a fairly rare event, so that we would not expect to “capture” this event by taking snapshots of every ten generations for just two runs, as illustrated in the BWM. In short, you agree with my co-tuning of mutation rate and generation size analysis, and that it will lead to evident o/p latching or quasi-latching under certain circumstances. under others, esp by making mutation rate per letter per member high enough,a nd making generation size large enough that the far-skirt members are more likely to be present, we will see that the rewarding of mere proximity without reference to function will lead to domination of the succeeding generations by multiple correct letter members, and thus very fast runs tot target. In an intermediate range, we should see more and more of reversion of letters to incorrect status,a s a double mutation substitutes another correct letter. The 1987 run seems to come from that intermediate range. 6] Chance of correct staying correct is 95%, and chance of correct changing to incorrect is 5%. (Remember, no latching) No EXPLICIT latching. Latching (and quasi-latching) as I have discussed for quite some time in this thread, and in the previous one -- and as J has also pointed out -- can be implicit, an effect of interacting factors. 7] the probability of incorrect changing to correct is 5% • 1/26 = 0.2%, or about 1/500 of the time. Therefore the probability of incorrect staying incorrect is 99.8% Thus, implicit quasi-latching,and effective latching in the case of appropriately co-tuned population scope and mutation rates. (Which can probably be done by trial and error looking for "good" results.) 8] the probability of a correct letter mutating to incorrect AND that child phrase being better than the parent are very small because it would take at least two or more incorrect to correct mutations to compensate and improve the fitness of the child Oops. This unfortunately neglects the case of the child simply being equal in [Hamming] distance to target as the parent, the latter being the champion from the previous generation. This raises the interesting issue of a tie between zero-change members and a double change member that substitutes a new correct letter for an old correct letter that has reverted. there may be a randomiser in the program [flip a coin and pick], or a line that selects the changed phrase by preference in such cases. It is in the case of the double-mutation with such substitution that we are most likely to see reversion behaviour. (My point of concern on this is that in the 1987 run, we see a further winking effect -- very rapid reversion to correct -- which seems odd if the letter is now simply on the long odds of getting back correct by mere chance. But, I am willing to not revert to the easiest [T2, Apollos variant] explanation for such, on grounds of taking testimony as so absent specific and strong grounds for rejecting it.) 8] The conclusion here is, then, that correct letters can mutate to incorrect but they very seldom survive because they weaken the fitness of the phrase too much. This is exactly why Joseph and I have spoken of two patterns of algorithms, those that explicitly latch, and those that implicitly latch. In short, your explicit binomial theorem based mathematical, probability analysis agrees with our narrative analysis based on programming patterns and population behaviour patterns. Indeed, you have confirmed that the implicit latching alternative is a viable mechanism for the o/p behaviour circa 1986. HOWEVER, YOU HAVE ALSO UNFORTUNATELY MANAGED AT THE OUTSET TO MISS THE POINT WE HAVE REPEATEDLY MADE HERE AND IN THE PREVIOUS THREAD, THAT THERE IS IN ADDITION TO EXPLICIT LATCHING, IMPLICIT LATCHING (AND QUASI-LATCHING). ______________ So, your analysis agrees with ours, but your conclusion as stated is unfortunately flawed because you missed the point that we have looked at BOTH explicit latching and implicit latching. Also, I again call attention to Apollos' code example that shows that EXPLICIT latching can be set up to mimic implicit quasi latching -- with reversions. So, can we agree that the Weasel program circa 1986 was probably a version of T3, with implicit latching, but in 1987, parameter changes would have made the latching shift to quasi-latching, without a tearaway streak to the target? GEM of TKIkairosfocus_{March 23, 2009
March
03
Mar
23
23
2009
12:35 AM
12
12
35
AM
PDT}

Hazel and Upright (and onlookers): First, Hazel, thanks for taking time to engage on substance, not rhetoric. (you have missed a key point in J's and my case, that we embrace not only explicit but internal latching or quasi-latching) but the analysis allows us to move the issue forward I believe. I believe the misunderstanding of our position is obviously inadvertent, not at all a calculated strawman. Upright, thanks for putting up a very telling wider analysis by Mr Berlinski. On points: 1] Hazel, 200: Some (notably Joseph and kairosfocus (GEM)) claim, based on the examples, that the BWM algorithm DID use latching (and was thus different than the video version), while others argue that the BWM algorithm and the video algorithm are the same Both Joseph and I are more nuanced than that. The observation is evident latching on the o/p circa 1986 contrasted with a run in 1987 that on video multiply and frequently flicks back (including a "winking" effect where reversion to correct seems to happen rather quickly). The issue we have had -- and J and I are not simply equivalent -- is explicit vs implicit latching, with the possibility of a quasi latching also in play. [J has stated that implicit latching is a most likely explanation, as a byproduct of the targetted search algorithm used as described by Mr Dawkins. I have pointed out that the core issue on Weasel is that it is targetted search that rewards non-functional configs, and that this obvioates any claims to being an exemplar of a BLIND watchmaker at worjk, ever since December last. As a secondary point I have pointed out to the evident o/p latching (as have many others over these 23 years) and have suggested that the most likely explanation is latching in the program, easiest as explicit, though also possibly implicit.] On Mr Dawkins' testimony as reported by Mr Elsberry [and I note, in light of a recent comment by a certain Dr Simmons, that "Mr" is strictly correct even in reference to one holding a PhD, save where disrespect is contextually evident; I have meant no disrespect], I have accepted it ar reasonable to address implicit latching. In that context, I have pointed out that once a per letter per member of a generation mutation rate and population size are set in a range that a significant fraction will be 0-change and thereafter decreasing fractions will have 1 or more changes, multiplied by a population size that makes the skirt unlikely to appear in the cases where doubly changed letters substitute so that one reverts while one advances, we will reasonably get the sort of pattern observed in the 1986 runs. In particular, I commented on the point that if double and triple etc mutations are common, i.e. there is a fairly large population, we will see quite fast runs to target -- instead of runs that are consistent with "good" runs for latching; which has median run length at 98 generations. It is credible that such co-tuning can be achieved with a modest number of test runs and intuitively appe3laing mutation rates. [The 5% rate I put up is from others, I believe tracing inter alia to Rob. It averages out at a bit more than 1 per member of the generation.] Next, I have had to point out that there is such a thing as a reasonable sample size in empirical contexts, that will be credibly typically representative of the population as a whole. And 200 - 300 in a context that we may safely presume means that the samples were selected for being typical of "good" runs, is beyond reasonable doubt, well within that range. 2] Upright, 202 and 205: Advent of the Head Monkey Thank you. This aptly captures the wider context of the issue, and sets up any serious discussion on the merits. I both agree with Mr Berlinski -- and yes, I know he holds a PhD (in I believe mathematics) -- and find him to have admirably presented the matter. 3] H, 200: I take “latching” to mean that once a correct letter appears in the correct slot, it can NOT mutate again, and will therefore stay correct for the duration of the program. This only describes a simplistic version of EXPLICIT latching; more sophisticated explicit latching is possible and implicit latching and quasi-/ imperfect- latching are also possible. By way of example, in the previous thread, Apollos presented code which will explicitly latch AND will have letters latched revert to incorrect status. (This means straight off that only credible code is actually demonstrative on the status of Weasel circa 1986.) I have spoken to both implicit [T3] and explicit [T2] latching in processing, and to evident latching of output. The T2 and T3 options are alternative explanations for the observed output. In light of the further statement of Mr Dawkins circa 2000 received at second hand, we have in the main discussed T3 in this thread. In short, the issue is somewhat mis-framed in the opening points of comment no. 200. 4] The question is whether he used the same program when he wrote the Blind Watchmaker (BWM). Actually, th sis the secondary question, the primary one being the rhetorical status of Weasel vs its true substantial import. I and others have objected that in context it serves to lend persuasive force tot he BLIND watchmaker thesis, while in fact it is an example of designed, targetted search that uses a toy example and simplifying assumptions that evade the force of the Hoylean challenge of getting TO shores of functionality before hill climbing based on differential functionality can be brought to bear. On the second level point, the issue is not whether the algorithm structure circa 1986 is different from that of 1987, but -- on the presumption of some version of a T3, implicit latching or quasi-latching algorithm from 1986 on -- whether parameters and o/p behaviour are credibly the same. In fact, early in the thread, Mr Dembski stated that such a shift could have material impact. I have held that there is good reason to infer that the printed excerpts circa 1986 are representative of what was thought to be "good" performance at that time, and that this behaviour shows strong latching on the o/p, up to runs to target consistent with that. By sharp contrast, circa 1987, just as strongly, the o/p does NOT latch, but shows frequent reversions, often with winking. I have concluded that the program circa 1987 is materially different form that circa 1986; which on T3 can be managed by changing two parameters: generation size and per letter per member mutation rates. It is probably noteworthy that an o/p that winks and runs fast but takes fairly long to get to target, is visually impressive. [ . . . ]kairosfocus_{March 23, 2009
March
03
Mar
23
23
2009
12:35 AM
12
12
35
AM
PDT}

Prev 1 … 3 4 5 6 7 … 12 Next